Journal of Power Sources 112 (2002) 98±104
Self-discharge of secondary lithium-ion graphite anodes
Chunsheng Wanga,*, Xiang-wu Zhanga, A. John Applebya,
Xiaole Chenb, Frank E. Littlec
a
Center for Electrochemical Systems and Hydrogen Research, Texas Engineering Experiment Station,
Texas A&M University, College Station, TX 77843-3402, USA
b
Department of Chemistry, Texas A&M University, College Station, TX 77843-3255, USA
c
Center for Space Power, Texas Engineering Experiment Station, Texas A&M University, College Station, TX 77843-3118, USA
Received 29 April 2002; accepted 10 June 2002
Abstract
A new method for measuring self-discharge of lithium-ion graphite anodes is reported. First, the anode lithium content is calculated from
the open-circuit potential as a function of open-circuit time (OCP versus OCT) using equilibrium potential±composition isotherms. Then, the
self-discharge rate is obtained from the differential of lithium content with respect to time on open circuit. The self-discharge rate measured by
this means after the ®rst charge±discharge cycle to 0.0 V versus Li/Li‡ is largely controlled by the growth of the solid electrolyte interphase
(SEI) ®lm, with some in¯uence from stage transformation.
# 2002 Elsevier Science B.V. All rights reserved.
Keywords: Self-discharge; Graphite anode; Li-ion batteries; Solid electrolyte interphase
1. Introduction
Lithium-ion batteries are now widely used for such civilian
applications as portable telephones and lap-top computers
because they offer signi®cant advantages in speci®c energy
and energy density over the state-of-the-art nickel systems.
However, for aerospace applications such as the outer planet
and solar probe (OP±SP) missions and geosynchronous earth
orbit (GEO) satellites, extended operating or calendar lives of
10±15 years are required. Generally, lithium-ion cells
undergo self-discharge. This is less signi®cant than that of
nickel±cadmium and nickel±metal hydride cells, but is still
both relatively rapid and temperature dependent. Recent
work shows that the self-discharge of LiCoO2- or LiNixMyO2-graphite cells is controlled by the progressive growth
of the solid electrochemical interphase (SEI) ®lm on the
anode, which consumes intercalated lithium [1]. Much of the
capacity loss in the graphite anode may be recovered by
recharging when a source of excess lithium is available in the
counter electrode. However, the entire initial capacity of the
graphite is not fully recoverable because the Li‡-conducting
SEI ®lm formed on individual graphite particle clusters is an
*
Corresponding author. Tel.: ‡1-979-845-0622/2033;
fax: ‡1-979-845-9287.
E-mail address: cswang@tamu.edu (C. Wang).
electronic insulator, and its continued growth will eventually
disintegrate the clusters. Hence, graphite anode self-discharge involves reversible and irreversible losses [2].
Usually, anode self-discharge as a function of open-circuit
time is measured by fully charging the anode, relaxing for
different open circuit times, then discharging at a low
current. However, the accuracy of this conventional method
is low. First, Li consumption due to SEI ®lm growth occurs
during both the time of relaxation and that of measurement of
charge retention. Hence, the total time on open circuit should
be much larger than that for charge retention measurement.
Accuracy therefore suffers since insuf®cient data can be
obtained. Secondly, the reaction overpotential increases with
open-circuit time due to continued SEI ®lm growth. Charge
retention depends on the charge and discharge rate used on
previous cycling [3]. Obtaining accurate data requires the use
of small charge and discharge currents whose values will be a
function of the previous history of relaxation. These currents
must both fully charge the anode and measure the real charge
retained. However, the use of low discharge currents
increases the discharge time, so charge retention measured
this way will diverge from accurate values.
In the present study, a new more accurate method has been
developed to measure the self-discharge rate of Johnson
Matthey (JM) 287 graphite anodes at 25 8C, whose selfdischarge mechanism is evaluated.
0378-7753/02/$ ± see front matter # 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 8 - 7 7 5 3 ( 0 2 ) 0 0 3 5 9 - 2
C. Wang et al. / Journal of Power Sources 112 (2002) 98±104
99
2. Experimental
2.3. Self-discharge measurement of JM 287 graphite
2.1. Electrode and cell preparation
After charging at 5.0 mA/g to 0.0 V, the graphite electrode
was disconnected, and the open-circuit potential of graphite
electrode as a function of time was monitored using the
Arbin cycler.
Electrodes were prepared from a mixture of 92 wt.% ballmilled Johnson Matthey (JM) 287 graphite powder (particle
size ca. 15 mm, BET area >30 m2/g) and 8 wt.% polyvinylidene ¯uoride between two nickel screen current collector
using 1-methy1-2-pyrrolidinone as solvent. After drying
overnight at 120 8C, the electrode was pressed into a sandwich structure with a geometric surface area of 2.0 cm2 and
weight of 40 mg. Electrochemical measurements were conducted in a three-electrode PTFE cell with two lithium foils
as counter and reference electrodes as in earlier work [4].
The electrolyte was 1.0 M lithium hexa¯uorophosphate
(LiPF6) in a 1:1:3 (v/v) ethylene carbonate (EC)±propylene
carbonate (PC)±dimethylcarbonate (DMC) mixture (High
Purity Lithium Battery Grade, Mitsubishi Chemical Company). All potentials given are versus the Li/Li‡ reference
electrode in this electrolyte. Charge (lithium insertion) and
discharge (lithium extraction) characteristics were measured between 0.0 and ‡1.5 V at constant current using
an Arbin Instruments (College Station, TX) automatic
battery cycler.
2.2. Li extraction equilibrium potential±composition
isotherm measurements
These were measured by galvanostatic intermittent titration (GITT) with microcurrent [4]. After ®ve charge±discharge cycles to stabilize the SEI ®lm, the anodes were
charged at 5 mA/g to 0.0 V, and then further charged potentiostatically at 0.0 V for 2.0 h. Following this, the anodes
were discharged under GITT conditions in a series of
intermittent discharges at 5 mA/g for 1.0 h, leaving the
electrode at open circuit for 2.0 h between each discharge.
The amount of extracted Li from graphite (m in Li1 mC6) in
each intermittent discharge was 0.0134 (…72 5†/26,800).
The open-circuit potential after 2.0 h relaxation as a function
of Li content was obtained. The open-circuit time between
successive discharge steps was determined to be rather
critical, since it must be suf®ciently long to establish equilibrium, but short enough so the self-discharge was not
signi®cant. To verify that a 2.0 h relaxation time was suf®cient to establish equilibrium, the characteristic potentials
(the potentials at the start and end of successive stage
transformations, i.e. the potential at the start and end of
each potential plateau) in the potential±composition isotherms measured using GITT were compared with the
characteristic potentials during open circuit in the selfdischarge measurement. The characteristic potentials were
the same in both cases, con®rming that potential±composition curves obtained using GITT represent equilibrium
potential±composition isotherms. In fact, GITT had been
successfully used to measure the reversible potential as a
function of Li content in graphite [5].
3. Results and discussion
3.1. Self-discharge of JM 287 graphite after the
first charging to 0.0 V
Fig. 1 shows the open-circuit potential (OCP or equilibrium potential)±composition isotherms (OCP versus x in
LixC6) measured by GITT during the sixth Li extraction
cycle after full Li insertion following the above procedure,
as well as the open-circuit potential as a function of time.
Both the OCP in the open-circuit potential±composition
isotherms (OCP versus x in LixC6) and the OCP pro®les
during open-circuit time in self-discharge measurement
show typical Li graphite intercalation compound characteristics, i.e ®ve stages for Li extraction from LiC6 in
the potential range below 0.3 V (Fig. 1b). These are
ascribed to ®ve continuous transitions at compositions
LiC6 ! LiC12 ! LiC18 ! LiC27 ! LiC36 ! LiC72 [5].
The three Li extraction potential plateaus at around 89,
126, and 218 mV are exactly the same as a previous study
[5]. As stated above, the characteristic potentials in the 2.0 h
open-circuit potential during GITT measurements are the
same as those of the open-circuit potential pro®le during
self-discharge, which con®rmed that potential±composition
curves obtained using GITT represent equilibrium potential±composition isotherms. To obtain the self-discharge
rate, the open-circuit potential (OCP) during self-discharge
measurement was expressed as the Li content in graphite
based on the OCP (or equilibrium potential)±composition
isotherms, which allows Li retention to be obtained as a
function of open-circuit time. In other words, according to
the OCP±composition isotherm curve, we can get the Li
content in graphite (x in LixC6) at a certain OCP. Therefore,
the OCPs in OCP versus OCT curves during the self-discharge measurements can be the change in Li content x, i.e.
the Li content x versus OCT curves can be obtained. For
example, after 85 h, the open-circuit potential in the selfdischarge potential pro®le is 0.09 V, corresponding to the
composition Li0.8C6. The Li retained as a function of time on
open-circuit time shown in Fig. 2. The instantaneous selfdischarge rate (Fig. 3) may be obtained by differentiating the
Li content in Fig. 2 with respect to open-circuit time. The
self-discharge rate in Fig. 3, shows three peaks in three
single phase regions and low values in the stage transformation regions. The intensities of the peaks in the single phase
region decreased with open-circuit time.
The self-discharge performance of graphite anodes may
be explained by the growth of the SEI ®lm on the graphite
100
C. Wang et al. / Journal of Power Sources 112 (2002) 98±104
Fig. 1. (a) Solid line: open-circuit potential of a JM 287 graphite anode as a function of time on open circuit after the first complete charge to 0.0 V at 5 mA/g
at 25 8C; dashed line: open-circuit potential (equilibrium potential)±composition isotherms measured at 25 8C using the GITT technique with successive
discharges at 5 mA/g for 1.0 h followed by 2.0 h on open circuit. (b) Enlarged part of (a) in the stage transformation region to show the characteristic stagetransformation potential.
particles. Li loss from graphite while on open-circuit results
from parasitic reactions between intercalated lithium and the
electrolyte. The ®rst explanation of the reduced ®lm growth
during self-discharge was given by Peled [6]. He ascribed it
to the progressively smaller electronic conductance of the
SEI ®lm as growth proceeds. The low-solubility SEI ®lm is
only a fraction of a micron in thickness [1], and consists of
an inner layer of largely inorganic material (Li2CO3, with
some LiF from the solute), and an outer rather porous layer
consisting of a graded mixture of inorganic material, alkyl
lithium carbonates, and polymeric hydrocarbons [1,7,8].
This may be characterized by a Li‡ concentration gradient
from 57 M (as Li2CO3) at the inner-to-outer layer boundary
to 1 M in the electrolyte. When completely formed, the inner
layer has the property of conducting some unsolvated Li‡,
but it probably cannot conduct solvated Li‡. The outer layer
has progressively smaller Li‡ conduction, and greater conduction of solvated Li‡, with increasing distance. The
mechanism of ®lm formation from the generic carbonate
ester solvent AO(CO)OB, which may or may not be cyclic
with AB joined, requires the transfer of one electron to
produce the ®rst alkyl carbonate ion product, AO(CO)O ,
plus the free radical B . This is followed by a second similar
electron transfer to form CO3 2 ‡ A . Combination of the
free radicals yields the polymeric material. The source of the
electrons is via discharge of Li atoms to Li‡ at the electrode
C. Wang et al. / Journal of Power Sources 112 (2002) 98±104
101
Fig. 2. Li retention in JM 287 graphite as a function of time on open circuit after the first complete charge to 0.0 V at 5 mA/g at 25 8C.
surface, but a means must be sought for their transfer from
the surface to the point where the carbonate ester or alkyl
carbonate ion becomes available. Electron tunneling
through the primary layer has been suggested as a probable
mechanism for transfer to that point [7]. However, some
means of transfer to the point at which reactants are available
is required. A further point is the nature of the reactant itself.
Carbonate ester coordinating Li‡ would be expected to be
less stable than the free solvent molecules, because of
electron displacement to the carbonate end of the molecule
via the induced dipole effect. In addition, the positive charge
on the solvated ion would represent a more favorable target
for electrons. However, the free radicals should readily be
able to accept electrons, giving carbanions. Thus, the reaction may proceed via a pool of free radicals carrying
electrons from the surface, where they react with solvated
lithium as in the following example (only one ligand is
shown, since the others detach after charge-transfer):
Li‰AO…CO†OBŠ‡ ‡ A ! Li‰AO…CO†OŠ ‡ A ‡ B
Following this step, the free radicals may polymerize, or
carry electrons from the surface. This explains how a free
radical pool occurs, since two are produced per electron
transfer. The insoluble product also forms in situ, and does
not require diffusion of positive and negative ions towards
each other. Further, the detached ligands form a pool for
solvation of the Li‡ produced by the electron transfer at the
graphite surface, which passes through the outer ®lm into the
electrolyte. Finally, slightly soluble alkali carbonate ion is
further oxidized in similar steps on the surface of the inner
Fig. 3. Self-discharge rate for JM 287 graphite anode obtained by differentiating Li content with respect to open-circuit time from the data in Fig. 2.
102
C. Wang et al. / Journal of Power Sources 112 (2002) 98±104
®lm. This provides a reasonably consistent model of ®lm
growth. The passivating inner layer controls the ¯ow of
electrons, which is the rate-determining step overall.
The Li loss process includes: (i) Li diffusion in the
graphite particle, (ii) movement of the Li phase boundaries
from the edges to the particle center, (iii) the ionic and
electronic mobilities of Li‡ and electrons in the SEI layer
[9]. As a ®rst approximation, the electronic mobility of Li‡
in the SEI layer is proportional to the SEI electronic conductance X [1]:
dx
kws
ˆ kX ˆ
dt
l0 ‡ Ax
(1)
where x is the number of moles of Li being reacting, w the
speci®c conductivity, s the interface area, l0 the initial layer
thickness, A and k are coef®cients. Eq. (1) indicates that if
the self-discharge rate is limited by the SEI electronic
conductance, then its rate should decrease with time on
open circuit [1]. Since the peak intensities for self-discharge
in the single phase region do indeed decrease with time on
open circuit (Fig. 3), this indicates that graphite anode selfdischarge is controlled by growth and thickness of the inner
SEI ®lm. The low self-discharge rate in the phase transformation region indicates the phase transformation barrier
decreases the self-discharge rate because the stage transformation has been con®rmed as the rate-limiting step of Li
extraction from graphite [4]. A difference in the self-discharge rate in the single- and two-phase regions has been
reported by Broussely et al. [1], who considered it due to
random scattering of the experimental results. A low selfdischarge rate in the two-phase region cannot be attributed to
a slow change of potential with time, because large composition changes occur over small ranges of potential in the ¯at
plateau region.
The advantage of the proposed method of measurement of
self-discharge is that it can give details of the self-discharge
rate within the different stage regions by simple measurement of the open-circuit potential as a function of time
in comparison with the equilibrium potential±composition
isotherms.
3.2. The self-discharge rate of JM 287 graphite after
five charge±discharge cycles
If the decreasing self-discharge rate is largely dependent
on the growth of the SEI ®lm, self-discharge should be much
lower after ®ve charge±discharge cycles, i.e. when the
passivating SEI ®lm is completely formed. Fig. 4 shows
the open-circuit potential of a graphite anode as a function of
time on open circuit after ®ve charge±discharge cycles. The
open-circuit potentials (or equilibrium potentials) versus Li
composition in graphite is also shown for comparison. The
time for the variation of open-circuit potential from 0.0 to
0.3 V after the sixth complete charge to 0.0 V is around
15.00 h (Fig. 4), which is almost three times longer than that
after the ®rst complete charge to 0.0 V (Fig. 1b). The amount
of Li retained after different times on open circuit was
obtained by the method described earlier and is shown in
Fig. 5a. The self-discharge rate was calculated by differentiating the Li content with respect to time on open circuit
(Fig. 5b). In contrast to the self-discharge rate after the ®rst
charge as shown in Fig. 3, the peak intensity of the rate of
Fig. 4. Solid line: open-circuit potential of JM 287 graphite anode at different open-circuit times after the sixth complete charge to 0.0 V at 5 mA/g at 25 8C.
Dashed line: open-circuit potential (Equilibrium potential)±composition isotherms measured at 25 8C using the GITT technique as in Fig. 1.
C. Wang et al. / Journal of Power Sources 112 (2002) 98±104
103
Fig. 5. (a) Li retention in JM 287 graphite after different times on open circuit; (b) self-discharge rate for JM 287 graphite anode after the sixth complete
charge to 0.0 V at 5 mA/g at 25 8C.
self-discharge in the single-phase region does not decrease
with time spent on open circuit and the self-discharge rate in
the x 1 (LixC6) single-phase region is even lower than that
in the following stage transformation (Li1 C6 ! Li0:5 C6 )
region. This behavior may be associated with (i) the coupled
processes of reductive solvent decomposition and dissolution of the largely soluble products in the electrolyte (insoluble products result in a loose and resistive SEI ®lm [10],
(ii) side reactions with electrolyte impurities, (iii) other
irreversible reactions, such as disintegration of graphite
particles after shrinkage during Li extraction.
The self-discharge rate of graphite electrode in 1 M
LiPF6 =EC ‡ PC ‡ DMC (1:1:3) electrolyte reported here
was much higher than that observed in practical batteries [1].
This may result from: (i) the high concentration of electrolyte PC, which is more reactive than other carbonates on
graphite, and (ii) the use of ¯ooded electrolytes.
Since the capacity of graphite anodes will decline due to
the disintegration resulting from repeated expansion/con-
traction and more resistive SEI ®lm growth between particles, the Li content in Li extraction equilibrium potential±
composition isotherms will decrease with increasing
charge±discharge cycles. However, the change in capacity
of the graphite anode during cycling will not in¯uence the
characteristic potentials in OCP (or equilibrium potential)±
composition isotherms. Hence, the reduction in capacity
means that the equilibrium potential±composition isotherms
used for self-discharge rate calculation should be measured
in the Li extraction cycle just before self-discharge measurement. For simpli®cation, the actual Li content in the equilibrium potential±composition isotherms at any cycle can be
approximately calculated by using a modi®ed coef®cient l
(LilxC6) in the standard equilibrium potential±LixC6 curve.
The coef®cient l can be approximately obtained by the ratio
of charge capacity at very low applied current before selfdischarge measurement to the theoretical capacity of graphite (372 mAh/g) if the Coulombic ef®ciency is considered
to be close to unity.
104
C. Wang et al. / Journal of Power Sources 112 (2002) 98±104
4. Conclusions
References
The self-discharge rate of Li-ion graphite anodes can be
calculated in a simple manner from the open-circuit potential as a function of time on open circuit using the equilibrium potential±composition isotherms. The self-discharge
rate measured by this means can give a detailed picture of
the self-discharge rate and Li content in the different stage
regions. The self-discharge rate of graphite anodes after the
®rst charge is controlled by the growth (i.e. effective thickness) of the SEI ®lm, although phase transformation from a
Li-rich to Li-poor stage decreases the self-discharge rate.
[1] M. Broussely, S. Herreyre, P. Biensan, P. Kasztejna, K. Nechev, R.J.
Staniewicz, J. Power Sources 97/98 (2001) 13.
[2] D. Gauyomard, J.M. Tarascon, Solid State Ionics 69 (1994) 222.
[3] P. Arora, R.E. White, M. Doyle, J. Electrochem. Soc. 145 (1998)
3647.
[4] C. Wang, I. Kakwan, A.J. Appleby, F.E. Little, J. Electroanal. Chem.
489 (2000) 55.
[5] T. Ohzuku, Y. Iwakoshi, K. Sawai, J. Electrochem. Soc. 140 (1993)
2490.
[6] E. Peled, J. Electrochem. Soc. 126 (1979) 2047.
[7] D. Aurbach, B. Markowsky, I. Weissman, E. Levi, Y. Ein-Eli,
Electrochim. Acta 45 (1999) 67.
[8] R. Yazami, Electrochim. Acta 45 (1999) 87.
[9] E. Peled, In: J.P. Gabano (Ed.), Lithium Batteries, Academic Press,
New York, 1983, pp. 43±72.
[10] S. Zhang, M.S. Ding, K. Xu, J. Allen, T.R. Jow, Electrochem. Solid
State Lett. 4 (2001) A208.
Acknowledgements
Financial support by the NASA-Glenn Research Center,
Cleveland, OH is gratefully acknowledged.